PPT

　　　　Copyright © 2006 Keio University

Networking Problems in Using Networking Problems in Using Quantum RepeatersQuantum RepeatersRodney Van MeterRodney Van MeterMAUI, 2009/4/16MAUI, 2009/4/16

rdv@ sfc.wide.ad.jprdv@ sfc.wide.ad.jphttp://www.sfc.keio.ac.jp/~rdv/http://www.sfc.keio.ac.jp/~rdv/

Assume a Quantum Computer Li ke Thi s. . .

I want to Bui l d a Di stri buted Quantum System Li ke Thi s

Laboratory-sized quantum multicomputer or transcontinental network, either one!

Repeater Protocol Stack

4

Van Meter et al., IEEE/ACM Trans. on Networking,Aug. 2009 (to appear), quant-ph:0705.4128

Physical Entanglement (PE)

Entanglement Control (EC)

Purification Control (PC)

Entang. Swapping Ctl (ESC)

Purification Control (PC)

Application

Distance=1

}

}Repeated atDifferent Distances

}End-to-End

Only quantum!

5

Outl i ne

•Two types of quantum networks•IPsec with QKD

•IPsec with QKD•US & European efforts•Open problems & plans

•Repeaters•Basic concepts•Our recent results•Open problems & plans

6

Two Types of Quantum Networks

UnentangledNetworks

EntangledNetworks

A B

C

E

G

H

7

Quantum Key Di stri buti on (QKD)

•Creates a share d, random se cre t between two nodes

•Uses physical effects to guarantee that key has not been observed

•Requires authenticated classical channel•Limited to <150km per hop

I Psec wi th QKD (ORF2008)

The DARPA Quantum Network

QKD Endpoint

QKD Switch

Eavesdropping

QKD Endpoint

QKD Switch

QKD EndpointQKD Switch

PrivateEnclave Private

Enclave

PrivateEnclave

BBN Harvard

BU

Dark MetroFiber

Lab Fiber

ConventionalEthernet

slide from Elliott, BBN

- SECOQC Prot ot ype pr i nc i pl el ayout

FOR81 m

LMU

Slide fromM. Peev, 2008

A Trusted repeater QKD-Network: Abstract Archi tecture (SECOQC, Europe)

QKD Access Node

QKD Core Node

Secrets Plane

QKD Access Node

QKD Core Node

VPN-greenSite 1

VPN-yellowSite 1 VPN-yellow

Site 2

Data Plane

Quantum Plane

Slide fromM. Peev, 2008

12

QKD wi th I Psec Pl ans

•Test over raw fiber, Yagami<->K2•Use key for one-time pad•Work w/ NEC, BBN & ITU to standardize•Write experimental I-D on IKE changes•Take to IETF in Hiroshima?

13

Outl i ne

•Two types of quantum networks•IPsec with QKD

•IPsec with QKD•US & European efforts•Open problems & plans

•Repeaters• Bas ic concepts

• Our recent res ults

• Open problems &plans

A B

C

E

G

H

14

Network Li nk Technol ogy (Qubus)

coherentoptical source(laser)

waveguide

homodynedetector

transceiverqubit innode 1

transceiverqubit innode 2

millimeters to kilometers

Munro, Nemoto, Spiller, New J. Phys. 7, 137 (2005)Ladd et al., NJP 8, 184 (2006)

15

Quantum Repeater Operati on:Entangl ement Swappi ng

Station 0 Station 1 Station 2

Bell State Measurement

Fidelity decreases; you must purify afterwards

Results must be communicated

16

Nested Entangl ement Swappi ng

17

Puri f i cati on

Station 0 Station 2

Results must be communicated (two-way?)

18

Repeater Protocol Stack

Van Meter et al., IEEE/ACM Trans. on Networking,Aug. 2009 (to appear), quant-ph:0705.4128

Physical Entanglement (PE)

Entanglement Control (EC)

Purification Control (PC)

Entang. Swapping Ctl (ESC)

Purification Control (PC)

Application

Distance=1}} Repeated at

Different Distances

} End-to-End

Only quantum!

19

Four-Hop Protocol I nteracti ons

Van Meter et al., IEEE/ACM Trans. on Networking,Aug. 2009 (to appear)

PEECPC

ESCPC

App

ESCPC

PEECPC

ESCPC

App

ESCPC

PEECPC

ESCPC

ESC

PEECPC

ESC

PEECPC

ESC

20

The Repeater’s Jobs

Entanglement swapping & purification, which require:

•A little bit of quantum communication•Quantum memory•Local quantum operations

(gates & measurements)•Lots of decision making

(both local and distributed)•Lots of classical communication

21

Entangl ement Pumpi ng

Ineffective w/ large fidelity difference

0.638

0.6380.72

0.6380.75

0.6380.77

0.6380.79

22

Symmetri c Puri f i cati on

Problems:Exact matching can require long waits.Not realistic whenmemory effects(decoherence)considered.Can deadlock ifresources are limited.

0.638

0.6380.72

0.638

0.638

0.797

X0.72

23

Greedy Puri f i cati on

Doesn’t wait foranything, useswhatever’s available.

Works well w/ largenumber of qubitsper repeater.

0.638

0.6380.72

0.638

0.638

0.757

X

24

Banded Puri f i cati on

Large gains in throughput.Moderate # qubits (5-50).Avoids deadlock.Realistic memory model.Simple to implement inreal time (even in HW).Probably not optimal,but probably close.

0.638

0.6380.72

0.638

0.638

0.797

X0.72

Divide fidelity spaceinto multiple bandse.g., above & below 0.70

25

Banded Puri f i cati on Performance

Van Meter et al., IEEE/ACM Trans. on Networking,Aug. 2009 (to appear), quant-ph:0705.4128

26

Banded Puri f i cati on Latency

Van Meter et al., IEEE/ACM Trans. on Networking,Aug. 2009 (to appear), quant-ph:0705.4128

27

Protocol Desi gn

28

Routi ng

Simple: use Dijkstra’s Shortest Path First.

...but we don’t yet know the cost metric.

D

F

A B

C

E

G

H

29

A Di f ferent Meani ng of “Whi ch Path?”

A B

C

DE

F

G

H

3 hops: ACGB4 hops: ACGHB ACEHB ADEHB ADFHB5 hops: ACEHGB ADEHGB ADECGB ADFHGB6 hops: ACECGHB7 hops: ADFHECGB ACEDCHGB

30

But What i s Di stance?

A B

C

DE

F

G

H

What if hops are not homogeneous?

Are 2n-1 hops, 2n hops,and 2n+1 hopssignificantly different?

31

How Do We Order These?

• How does number of links matter?

• Does number of w e a k links matter?

• Does position of weak link matter?

• Is cost ?a d d it iv e

• At this logical level,is this technology-independent?

32

Other Probl ems

• Defining swap points• Static or dynamic?• Avoiding leapfrog• Avoiding deadlock• Minimizing waits for

classical messages

33

Other Probl ems

Partial messaging sequence

Can this be made more efficient?

Due to memory degradation, gains will be better than linear

34

Leapfrog

Station 0 Station 1 Station 2 Station 3

35

Resource Management (QoS?)

A B

C

D

A<->B & C<->Dwant to talk.

Remember, it’s a distributed computation.

Worse, fragile quantum memory means thereis a hard real time component.

==>requires circuit switching???(bottleneck likely is memory per node)

36

Open Repeater Probl ems

•Well, repeater HW doesn’t work yet...–Sims of “weak links” mostly functional–Establishing swapping points–More dynamic behavior–Non-power-of-two hops–Finish & publish protocol state machine

37

Open Compl ex Network Probl ems

•Coding partially done–Using graphviz file format–Routing not done–Workload generator needs work–QoS / resource allocation not implemented

•Visualization of networks•Investigate graph states & quantum network coding•More detailed workload definition

38

Mi l estones for JSPS

•Define a cost metric(figure out if it’s additive!)

•Define a path selection algorithm•Define test cases•Simulate that set of test cases•Extend to topologically complex networks•Create static visualizations

39

Food for Thought

•When will first S cie nce or Nature paper appear us ing a quantum computer, but not about the quantum computer?

•That is, when will a quantum computer do science, rather than be science?

•Answers from quantum researchers range from “less than five years” to “more than forty years”

Copyright © 2006 Keio University 　｜　　40

Thanks

Thanks to Thaddeus Ladd, Bill Munro and Kae Nemoto (coauthors on much of this work), as well as Austin Fowler, Jim Harrington, Kohei Itoh, Agung Trisetyarso, Byung-Soo Choi, Shota Nagayama, and Takahiko Satoh

And funding from NICT, MEXT, NSF, the Mori Fund at Keio, and now J S P S for funding.

AQUA: Advanci ng Quantum Archi tecture

情報は物理である：情報は物理である：Information is physicalInformation is physical

http://www.sfc.wide.ad.jp/aqua/http://www.sfc.wide.ad.jp/aqua/